C1000-185 · Review mode

After prompt-tuning a generative model, accuracy is high but generated text is awkward and ungrammatical. Which metric and remediation pair best addresses the fluency issue?

• AA. Use perplexity to measure fluency and consider further language-model fine-tuning (or rewards from a grammar-focused critic) to improve it
• BB. Use the BLEU score and reduce the size of the training dataset
• CC. Use accuracy and shorten the maximum output length
• DD. Disable evaluation